Infrared Eye Image Classification: A Deep Learning Approach for Enhanced Eye State Detection

Applications, ranging from detection of fatigue to glance tracking for medical diagnosis, are highly reliant on accurate classification of states, open or closed. This work creates a more robust and efficient classifier based on deep learning using a complete set of IR eye images in the MRL Eye Dataset. Under different lighting conditions and varied modifications in devices, the architecture CNN was used to train the model in detecting eye states. Through much testing and optimization, the model achieved brilliant validation accuracy of 98%, which probably deepened its ability to generalize in different test settings. Techniques used to improve the performance and reduce overfitting include data augmentation, dropout regularization, and fine-tuning. The approach proposed here provides a scalable solution for many real-world applications requiring high levels of dependability and accuracy and really underscores the value of IR imaging for classifications of conditions in the eye. This work emphasizes the scope of possibilities whereby one might push the boundaries of eye state recognition systems by the incorporation of such domain-specific datasets with advanced deep learning methods.